Build, Don't Play: Self-Improving LLM Research Agents Develop a Hearthstone Battlegrounds AI That Rivals Top Players

A research effort dubbed AutoTavern reports that self-improving LLM research agents built a Hearthstone Battlegrounds AI that holds its own against top human players, according to the team's write-up of the work. The project targets Hearthstone Battlegrounds, an eight-player auto-battler with stochastic shops and battles, hidden information and nearly a thousand interacting cards, and frames the result as a test of whether LLM agents can carry a hard real-world problem from raw data to a competitive system.

Can LLM agents carry a hard real-world problem from raw data to a system that holds its own against people? We study this in Hearthstone Battlegrounds, an eight-player auto-battler with stochastic shops and battles, hidden information and nearly a thousand interacting cards. We present AutoTavern, a